In a recent study from Accenture, 97 percent of the senior executives surveyed see great value in creating a data-centric culture, yet only 17 percent had actually gone on to implement a big data strategy in their business.
David Simchi-Levi, Best Selling Author and Award Winning Professor of Supply Chain Management, examines this incongruity and reveals how business leaders can transform their company into a data-driven enterprise.
What does Leo Tolstoy’s War and Peace have to do with big data?
Leo Tolstoy’s War and Peace is a 1,225-page critically-acclaimed story of life, love, and loss. For even the most avid bookworms the thought of reading this lengthy tome from start to finish is more of a millstone than a milestone. It’s a book that many people talk about but very few have actually read or finished.
The big data analytics narrative is similar. While this ground-breaking technology is firing the imagination of Supply Chain and Revenue Management professionals everywhere, most of them seem to be ‘talking the talk’ rather than actually ‘walking the walk’.
Indeed, if you examined the statistics you could be forgiven for thinking that big data had been implemented on a global scale. Take a recent report commissioned by SCM World [i] for example. It revealed that nearly two-thirds of executives think that big data analytics has the power to revolutionise supply chain management. And another study commissioned by the global management consulting firm, McKinsey & Company [ii], found that Supply Chain professionals who make the effort to embed a data-driven culture in their businesses are five percent more profitable than their rivals.
The missing piece
Curiously, very few senior executives seem to be missing the key aspect in making a transition to data-driven supply chain: creating a culture to support it. Recent figures from global professional services giant, Accenture [iii], for instance, make for staggering reading. The survey which interviewed over a thousand senior executives highlighted, that while 97 percent of them saw great value in inculcating a data-driven culture, astoundingly, only 17 percent had actually gone on to implement a big data strategy in their business.
So, if today’s captains of the industry broadly agree that big data analytics can transform supply chains, why have so few of them actually utilised big data in their organisations? Why aren’t there more data-driven enterprises? Surely, in a highly connected global workplace where information needs to flow at lightning speed, this disruptive and vitally important technology can lead to better and more efficient decision making?
I don’t wish to dwell too much on an apparent lack of big data take-up in industry, as that’s not the point of this blog post. But to quote another literary heavyweight, maybe H.P Lovecraft’s assertion that ‘the strongest kind of fear is the fear of the unknown’ [iv] may help explain why those overseeing global supply chains are so reticent to fully utilise big data analytics.
However, that said, there are many success stories – enterprises who have seen the bigger picture and have shaped their culture to embrace their evolved data-driven supply chain.
How Rue La La Increased Revenue by 11 Percent with Big Data
This Boston-based e-retailer specialises in flash sales, an e-commerce business model in which a website offers a single product for sale for a period of 48 hours. The industry, which started in mid-2000 is flourishing. Currently, it’s growing at a rate of 17 percent per year, which is much faster than a typical online retail company. [v]
Rue La La, however, realised that it could do better, and enlisted MIT’s help in implementing a set of data-centric tools, systems, and technologies to its boost its sales performance.
At this point, let me first say a little about Rue La La’s pricing strategy. It uses a Cost Plus pricing model, which focuses exclusively on profit margin.
But there’s the rub. In the world of flash sales, a successful sales campaign is judged on a salesperson’s ability to shift all the stock in the first sales event, that is, the first time Rue La La exposes a product to the market. A related challenge is associated with Rue La La selling an eclectic mix of products, ranging from designer watches to women’s running shoes. This implies that often the sales team has little or no experience selling a specific product. But with no historical data to call upon, how can effective decisions on pricing be made? [vi]
We, therefore, decided to analyse the effectiveness of its Cost Plus model by plotting a graph which measured the percentage of goods sold in the first sale against the frequency in which that was achieved.
Our research was quite revealing. Dividing the products into separate departmental categories, Iwe discovered that for one department, in 62 percent of cases the company sold out, which suggested that maybe the products had been sold too cheaply. [vii]
At the other end of the spectrum, the research highlighted that for another department, in 22 percent of cases, Rue La La sold almost nothing, which suggested that the price was perhaps too high. [viii]
Download David's Research Summary on Price Optimization here:
In meeting the challenge of pricing the product correctly, our research revealed one key point. Rue La La’s issues had nothing to do with inventory. Instead, the key to maximising revenue, profit and market share, centred on implementing an effective pricing strategy and embedding a data-driven culture.
Therefore, we developed a two-step approach which focused specifically on:
- Demand Forecasting
- Price Optimisation
1. Demand Forecasting [ix]
I’ve already talked about the challenge of overcoming a lack of historical data for first-time sale products. But, arguably a more complex issue is when an item sells out in just a few hours. The problem with a stock-out is that while I have plenty to go on from a sales perspective, I have no data in regard to demand.
So how did we rectify this issue? Well, we used machine learning techniques to reverse engineer sales information to generate powerful market demand data.
Firstly, we compared all the products that did not stock out with those that did. Secondly, we combined this internal data with external data from other sources.
By combining both sets of data we tested many different forecasting techniques. Our research revealed that using Regression Trees, which compare the price of the product to all competing products, was the most effective way of forecasting customer demand.
While on one level, a Regression Tree is an advanced analytic tool, let’s look at the bigger picture. These Trees are really a collection of rules, which if followed, tell us a story about marketing demand. Follow the narrative in one direction and it gives you one prediction, take another path and you’ll get another insight into customer demand.
But why are Regression Trees so effective? Well, in the world of Flash Sales, if I am selling a women’s running shoe, the Regression Tree will identify a subset of styles that I have sold before that are “similar” to the product I am currently selling and will learn from these styles.
Another advantage of Regression Trees is that it can take into account that the price/demand relationship of some products is non-monotonic, particularly when brand quality comes into play. Take a Ralph Lauren jacket for instance. Often, the higher the price, the higher the demand, which is something that a Linear Regression, the standard forecasting model most retailers apply, would not be able to process.
- Price Optimisation [x]
With the Regression Tree in place, the next challenge was to integrate it as part of a price optimisation process.
Customer demand for a specific product, of course, depends not only on its price but also the price of a competing product. So, for the pricing strategy technology tool to work, it needed to be able to price every competing product at the same time.
The second big hurdle that we had to overcome was not only to incorporate the Regression Trees into a price optimisation tool but in doing so ensure that the analytics were deeply embedded into the very heart of the company to create a data-driven culture. Therefore, we installed our Pricing Decision Tool directly into Rue La La’s IT infrastructure, which meant no information silos as everyone had access to the same data.
Implementing a Data-Driven Culture [xi]
Most large enterprises use highly developed Enterprise Resource Planning (ERP) systems to collect store, manage and interpret data. But in today’s fast-moving world, it is bespoke algorithms that are driving the big data revolution.
Take Rue La La’s ERP system for instance. Every night our Pricing Decision Tool pulls data from the Extraction, Transformation, and Loading (ETL) database. It then filters the information before placing it in the Retail Price Optimiser Database. From there, the data is further analysed by a collection of statistical tools including ‘R’ to generate the Regression Tree. At the same time, Inventory Information is applied to the Regression Tree to make sure that prediction is consistent with supply. After the final set of Regression Trees are transferred into the optimisation model, it takes between 45 minutes and one hour for the model to generate prices for thousands of different products.
When the merchants arrive in the office the next morning, they study our price recommendations and they typically approve around 95 percent of the recommended prices.
Measuring Data’s Effectiveness - Impact on Revenue and Market Share
But just how effective is our technology as a revenue generation tool? In September 2014, we decided to run a battery of offline analysis in an attempt to estimate the value of the Price Optimisation Tool.
We identified 6,000 styles where our technology had recommended a price increase. The reason we focused on price increase is that executives at Rue La La were concerned that while a price increase may increase revenue, it may decrease market share.
We then divided the 6,000 styles into five separate categories based on price. We split the styles in each price category into two groups, equally divided and let our algorithm generate the price for one group that include half of the products (styles), while Rue La La’s Cost Plus model was used to generate prices for the other half of products in that price category.
Astonishingly, our findings revealed that, across the five product categories, when our Price Optimisation Model was used it increased Rue La La’s revenue by an average of 11 percent without reducing market share.
Perhaps it serves as an important reminder that in our data-driven world, where the amount of information collected annually is increasing at a rate of 59 percent each year, it’s not the data that counts, but how we use it.
[i] Computer World, By Antonia Renner, Overcoming 5 Major Supply Chain Challenges with Big Data Analytics (paragraph one)
Big Data Analytics and the Evolution of the Supply Chain
By Andrew Armstrong
Computer World, By Antonia Renner, Overcoming 5 Major Supply Chain Challenges with Big Data Analytics (paragraph one)
[iv] H. P Lovecraft quote
[v] Taken from MiT Sloan Executive Education webinar ‘The New Frontier in Price Optimisation’ presented by Professor David Simchi-Levi
[vii] Taken from MiT Sloan Executive Education webinar ‘The New Frontier in Price Optimisation’ presented by Professor David Simchi-Levi
[viii] Taken from MiT Sloan Executive Education webinar ‘The New Frontier in Price Optimisation’ presented by Professor David Simchi-Levi
[ix] Taken from MiT Sloan Executive Education webinar ‘The New Frontier in Price Optimisation’ presented by Professor David Simchi-Levi
[x] Taken from MiT Sloan Executive Education webinar ‘The New Frontier in Price Optimisation’ presented by Professor David Simchi-Levi
[xi] Taken from MiT Sloan Executive Education webinar ‘The New Frontier in Price Optimisation’ presented by Professor David Simchi-Levi